Adaptive Discretization for Event PredicTion (ADEPT).

Recently developed survival analysis methods improve upon existing approaches by predicting the probability of event occurrence in each of a number pre-specified (discrete) time intervals. By avoiding placing strong parametric assumptions on the event density, this approach tends to improve prediction performance, particularly when data are plentiful. However, in clinical settings with limited available data, it is often preferable to judiciously partition the event time space into a limited number of intervals well suited to the prediction task at hand. In this work, we develop Adaptive Discretization for Event PredicTion (ADEPT) to learn from data a set of cut points defining such a partition. We show that in two simulated datasets, we are able to recover intervals that match the underlying generative model. We then demonstrate improved prediction performance on three real-world observational datasets, including a large, newly harmonized stroke risk prediction dataset. Finally, we argue that our approach facilitates clinical decision-making by suggesting time intervals that are most appropriate for each task, in the sense that they facilitate more accurate risk prediction.

Evaluating factors associated with telehealth appropriateness in outpatient rheumatoid arthritis encounters using the Encounter Appropriateness Score for You (EASY).

OBJECTIVE: Telehealth has been proposed as a safe and effective alternative to in-person care for rheumatoid arthritis (RA). The purpose of this study was to evaluate factors associated with telehealth appropriateness in outpatient RA encounters. METHODS: A prospective cohort study (1/1/21-8/31/21) was conducted using electronic health record data from outpatient RA encounters in a single academic rheumatology practice. Rheumatology providers rated the telehealth appropriateness of their own encounters using the Encounter Appropriateness Score for You (EASY) immediately following each encounter. Robust Poisson regression with Generalized Estimating Equations (GEE) modeling was used to evaluate the association of telehealth appropriateness with patient demographics, RA clinical characteristics, comorbid non-inflammatory causes of joint pain, previous and current encounter characteristics, and provider characteristics. RESULTS: During the study period, 1,823 outpatient encounters with 1,177 unique RA patients received an EASY score from 25 rheumatology providers. In the final multivariate model [Relative Risk (95% Confidence Interval)], factors associated with increased telehealth appropriateness included higher average provider preference for telehealth in prior encounters [1.26 (1.21-1.31)], telehealth as the current encounter modality [2.27 (1.95-2.64)], and increased patient age [1.05 (1.01-1.09)]. Factors associated with decreased telehealth appropriateness included moderate [0.81 (0.68-0.96)] and high [0.57 (0.46-0.70)] RA disease activity and if the previous encounter were conducted via telehealth [0.83 (0.73-0.95)]. CONCLUSION: In this study, telehealth appropriateness was most associated with provider preference, the current and previous encounter modality, and RA disease activity. Other factors like patient demographics, RA medications, and comorbid non-inflammatory causes of joint pain were not associated with telehealth appropriateness.

Designing an Implementable Clinical Prediction Model for Near-Term Mortality and Long-Term Survival in Patients on Maintenance Hemodialysis.

RATIONALE & OBJECTIVE: The life expectancy of patients treated with maintenance hemodialysis (MHD) is heterogeneous. Knowledge of life-expectancy may focus care decisions on near-term versus long-term goals. The current tools are limited and focus on near-term mortality. Here, we develop and assess potential utility for predicting near-term mortality and long-term survival on MHD. STUDY DESIGN: Predictive modeling study. SETTING & PARTICIPANTS: 42,351 patients contributing 997,381 patient months over 11 years, abstracted from the electronic health record (EHR) system of midsize, nonprofit dialysis providers. NEW PREDICTORS & ESTABLISHED PREDICTORS: Demographics, laboratory results, vital signs, and service utilization data available within dialysis EHR. OUTCOME: For each patient month, we ascertained death within the next 6 months (ie, near-term mortality) and survival over more than 5 years during receipt of MHD or after kidney transplantation (ie, long-term survival). ANALYTICAL APPROACH: We used least absolute shrinkage and selection operator logistic regression and gradient-boosting machines to predict each outcome. We compared these to time-to-event models spanning both time horizons. We explored the performance of decision rules at different cut points. RESULTS: All models achieved an area under the receiver operator characteristic curve of≥0.80 and optimal calibration metrics in the test set. The long-term survival models had significantly better performance than the near-term mortality models. The time-to-event models performed similarly to binary models. Applying different cut points spanning from the 1st to 90th percentile of the predictions, a positive predictive value (PPV) of 54% could be achieved for near-term mortality, but with poor sensitivity of 6%. A PPV of 71% could be achieved for long-term survival with a sensitivity of 67%. LIMITATIONS: The retrospective models would need to be prospectively validated before they could be appropriately used as clinical decision aids. CONCLUSIONS: A model built with readily available clinical variables to support easy implementation can predict clinically important life expectancy thresholds and shows promise as a clinical decision support tool for patients on MHD. Predicting long-term survival has better decision rule performance than predicting near-term mortality. PLAIN-LANGUAGE SUMMARY: Clinical prediction models (CPMs) are not widely used for patients undergoing maintenance hemodialysis (MHD). Although a variety of CPMs have been reported in the literature, many of these were not well-designed to be easily implementable. We consider the performance of an implementable CPM for both near-term mortality and long-term survival for patients undergoing MHD. Both near-term and long-term models have similar predictive performance, but the long-term models have greater clinical utility. We further consider how the differential performance of predicting over different time horizons may be used to impact clinical decision making. Although predictive modeling is not regularly used for MHD patients, such tools may help promote individualized care planning and foster shared decision making.

Evaluating Provider Perceptions of Telehealth Utility in Outpatient Rheumatology Telehealth Encounters.

OBJECTIVE: This study aims to explore the factors associated with rheumatology providers' perceptions of telehealth utility in real-world telehealth encounters. METHODS: From September 14, 2020 to January 31, 2021, 6 providers at an academic medical center rated their telehealth visits according to perceived utility in making treatment decisions using the following Telehealth Utility Score (TUS) (1 = very low utility to 5 = very high utility). Modified Poisson regression models were used to assess the association between TUS scores and encounter diagnoses, disease activity measures, and immunomodulatory therapy changes during the encounter. RESULTS: A total of 481 telehealth encounters were examined, of which 191 (39.7%) were rated as "low telehealth utility" (TUS 1-3) and 290 (60.3%) were rated as "high telehealth utility" (TUS 4-5). Encounters with a diagnosis of inflammatory arthritis were significantly less likely to be rated as high telehealth utility (adjusted relative risk [aRR], 0.8061; p = 0.004), especially in those with a concurrent noninflammatory musculoskeletal diagnosis (aRR, 0.54; p = 0.006). Other factors significantly associated with low telehealth utility included higher disease activity according to current and prior RAPID3 scores (aRR, 0.87 and aRR, 0.89, respectively; p < 0.001) and provider global scores (aRR, 0.83; p < 0.001), as well as an increase in immunomodulatory therapy (aRR, 0.70; p = 0.015). CONCLUSIONS: Provider perceptions of telehealth utility in real-world encounters are significantly associated with patient diagnoses, current and prior disease activity, and the need for changes in immunomodulatory therapy. These findings inform efforts to optimize the appropriate utilization of telehealth in rheumatology.

Epigenetic and transcriptional responses in circulating leukocytes are associated with future decompensation during SARS-CoV-2 infection.

To elucidate host response elements that define impending decompensation during SARS-CoV-2 infection, we enrolled subjects hospitalized with COVID-19 who were matched for disease severity and comorbidities at the time of admission. We performed combined single-cell RNA sequencing (scRNA-seq) and single-cell assay for transposase-accessible chromatin using sequencing (scATAC-seq) on peripheral blood mononuclear cells (PBMCs) at admission and compared subjects who improved from their moderate disease with those who later clinically decompensated and required invasive mechanical ventilation or died. Chromatin accessibility and transcriptomic immune profiles were markedly altered between the two groups, with strong signals in CD4+ T cells, inflammatory T cells, dendritic cells, and NK cells. Multiomic signature scores at admission were tightly associated with future clinical deterioration (auROC 1.0). Epigenetic and transcriptional changes in PBMCs reveal early, broad immune dysregulation before typical clinical signs of decompensation are apparent and thus may act as biomarkers to predict future severity in COVID-19.

Trans-Balance: Reducing demographic disparity for prediction models in the presence of class imbalance.

INTRODUCTION: Risk prediction, including early disease detection, prevention, and intervention, is essential to precision medicine. However, systematic bias in risk estimation caused by heterogeneity across different demographic groups can lead to inappropriate or misinformed treatment decisions. In addition, low incidence (class-imbalance) outcomes negatively impact the classification performance of many standard learning algorithms which further exacerbates the racial disparity issues. Therefore, it is crucial to improve the performance of statistical and machine learning models in underrepresented populations in the presence of heavy class imbalance. METHOD: To address demographic disparity in the presence of class imbalance, we develop a novel framework, Trans-Balance, by leveraging recent advances in imbalance learning, transfer learning, and federated learning. We consider a practical setting where data from multiple sites are stored locally under privacy constraints. RESULTS: We show that the proposed Trans-Balance framework improves upon existing approaches by explicitly accounting for heterogeneity across demographic subgroups and cohorts. We demonstrate the feasibility and validity of our methods through numerical experiments and a real application to a multi-cohort study with data from participants of four large, NIH-funded cohorts for stroke risk prediction. CONCLUSION: Our findings indicate that the Trans-Balance approach significantly improves predictive performance, especially in scenarios marked by severe class imbalance and demographic disparity. Given its versatility and effectiveness, Trans-Balance offers a valuable contribution to enhancing risk prediction in biomedical research and related fields.

Encounter Appropriateness Score for You Model: Development and Pilot Implementation of a Predictive Model to Identify Visits Appropriate for Telehealth in Rheumatology.

OBJECTIVE: We aimed to develop a decision-making tool to predict telehealth appropriateness for future rheumatology visits and expand telehealth care access. METHODS: The model was developed using the Encounter Appropriateness Score for You (EASY) and electronic health record data at a single academic rheumatology practice from January 1, 2021, to December 31, 2021. The EASY model is a logistic regression model that includes encounter characteristics, patient sociodemographic and clinical characteristics, and provider characteristics. The goal of pilot implementation was to determine if model recommendations align with provider preferences and influence telehealth scheduling. Four providers were presented with future encounters that the model identified as candidates for a change in encounter modality (true changes), along with an equal number of artificial (false) recommendations. Providers and patients could accept or reject proposed changes. RESULTS: The model performs well, with an area under the curve from 0.831 to 0.855 in 21,679 encounters across multiple validation sets. Covariates that contributed most to model performance were provider preference for and frequency of telehealth encounters. Other significant contributors included encounter characteristics (current scheduled encounter modality) and patient factors (age, Routine Assessment of Patient Index Data 3 scores, diagnoses, and medications). The pilot included 201 encounters. Providers were more likely to agree with true versus artificial recommendations (Cohen's κ = 0.45, P < 0.001), and the model increased the number of appropriate telehealth visits. CONCLUSION: The EASY model accurately identifies future visits that are appropriate for telehealth. This tool can support shared decision-making between patients and providers in deciding the most appropriate follow-up encounter modality.

Translating ethical and quality principles for the effective, safe and fair development, deployment and use of artificial intelligence technologies in healthcare.

OBJECTIVE: The complexity and rapid pace of development of algorithmic technologies pose challenges for their regulation and oversight in healthcare settings. We sought to improve our institution's approach to evaluation and governance of algorithmic technologies used in clinical care and operations by creating an Implementation Guide that standardizes evaluation criteria so that local oversight is performed in an objective fashion. MATERIALS AND METHODS: Building on a framework that applies key ethical and quality principles (clinical value and safety, fairness and equity, usability and adoption, transparency and accountability, and regulatory compliance), we created concrete guidelines for evaluating algorithmic technologies at our institution. RESULTS: An Implementation Guide articulates evaluation criteria used during review of algorithmic technologies and details what evidence supports the implementation of ethical and quality principles for trustworthy health AI. Application of the processes described in the Implementation Guide can lead to algorithms that are safer as well as more effective, fair, and equitable upon implementation, as illustrated through 4 examples of technologies at different phases of the algorithmic lifecycle that underwent evaluation at our academic medical center. DISCUSSION: By providing clear descriptions/definitions of evaluation criteria and embedding them within standardized processes, we streamlined oversight processes and educated communities using and developing algorithmic technologies within our institution. CONCLUSIONS: We developed a scalable, adaptable framework for translating principles into evaluation criteria and specific requirements that support trustworthy implementation of algorithmic technologies in patient care and healthcare operations.

Development and Optimization of a Subtraction-Normalized Immunocyte Profiling Signature for Prostate Cancer Active Surveillance Risk Stratification.

PURPOSE: Less invasive decision support tools are desperately needed to identify occult high-risk disease in men with prostate cancer (PCa) on active surveillance (AS). For a variety of reasons, many men on AS with low- or intermediate-risk disease forgo the necessary repeat surveillance biopsies needed to identify potentially higher-risk PCa. Here, we describe the development of a blood-based immunocyte transcriptomic signature to identify men harboring occult aggressive PCa. We then validate it on a biopsy-positive population with the goal of identifying men who should not be on AS and confirm those men with indolent disease who can safely remain on AS. This model uses subtraction-normalized immunocyte transcriptomic profiles to risk-stratify men with PCa who could be candidates for AS. MATERIALS AND METHODS: Men were eligible for enrollment in the study if they were determined by their physician to have a risk profile that warranted prostate biopsy. Both training (n = 1017) and validation cohort (n = 1198) populations had blood samples drawn coincident to their prostate biopsy. Purified CD2+ and CD14+ immune cells were obtained from peripheral blood mononuclear cells, and RNA was extracted and sequenced. To avoid overfitting and unnecessary complexity, a regularized regression model was built on the training cohort to predict PCa aggressiveness based on the National Comprehensive Cancer Network PCa guidelines. This model was then validated on an independent cohort of biopsy-positive men only, using National Comprehensive Cancer Network unfavorable intermediate risk and worse as an aggressiveness outcome, identifying patients who were not appropriate for AS. RESULTS: The best final model for the AS setting was obtained by combining an immunocyte transcriptomic profile based on 2 cell types with PSA density and age, reaching an AUC of 0.73 (95% CI: 0.69-0.77). The model significantly outperforms (P < .001) PSA density as a biomarker, which has an AUC of 0.69 (95% CI: 0.65-0.73). This model yields an individualized patient risk score with 90% negative predictive value and 50% positive predictive value. CONCLUSIONS: While further validation in an intended-use cohort is needed, the immunocyte transcriptomic model offers a promising tool for risk stratification of individual patients who are being considered for AS.

Host-response transcriptional biomarkers accurately discriminate bacterial and viral infections of global relevance.

Diagnostic limitations challenge management of clinically indistinguishable acute infectious illness globally. Gene expression classification models show great promise distinguishing causes of fever. We generated transcriptional data for a 294-participant (USA, Sri Lanka) discovery cohort with adjudicated viral or bacterial infections of diverse etiology or non-infectious disease mimics. We then derived and cross-validated gene expression classifiers including: 1) a single model to distinguish bacterial vs. viral (Global Fever-Bacterial/Viral [GF-B/V]) and 2) a two-model system to discriminate bacterial and viral in the context of noninfection (Global Fever-Bacterial/Viral/Non-infectious [GF-B/V/N]). We then translated to a multiplex RT-PCR assay and independent validation involved 101 participants (USA, Sri Lanka, Australia, Cambodia, Tanzania). The GF-B/V model discriminated bacterial from viral infection in the discovery cohort an area under the receiver operator curve (AUROC) of 0.93. Validation in an independent cohort demonstrated the GF-B/V model had an AUROC of 0.84 (95% CI 0.76-0.90) with overall accuracy of 81.6% (95% CI 72.7-88.5). Performance did not vary with age, demographics, or site. Host transcriptional response diagnostics distinguish bacterial and viral illness across global sites with diverse endemic pathogens.

A Deep-Learning Algorithm to Predict Short-Term Progression to Geographic Atrophy on Spectral-Domain Optical Coherence Tomography.

Importance: The identification of patients at risk of progressing from intermediate age-related macular degeneration (iAMD) to geographic atrophy (GA) is essential for clinical trials aimed at preventing disease progression. DeepGAze is a fully automated and accurate convolutional neural network-based deep learning algorithm for predicting progression from iAMD to GA within 1 year from spectral-domain optical coherence tomography (SD-OCT) scans. Objective: To develop a deep-learning algorithm based on volumetric SD-OCT scans to predict the progression from iAMD to GA during the year following the scan. Design, Setting, and Participants: This retrospective cohort study included participants with iAMD at baseline and who either progressed or did not progress to GA within the subsequent 13 months. Participants were included from centers in 4 US states. Data set 1 included patients from the Age-Related Eye Disease Study 2 AREDS2 (Ancillary Spectral-Domain Optical Coherence Tomography) A2A study (July 2008 to August 2015). Data sets 2 and 3 included patients with imaging taken in routine clinical care at a tertiary referral center and associated satellites between January 2013 and January 2023. The stored imaging data were retrieved for the purpose of this study from July 1, 2022, to February 1, 2023. Data were analyzed from May 2021 to July 2023. Exposure: A position-aware convolutional neural network with proactive pseudointervention was trained and cross-validated on Bioptigen SD-OCT volumes (data set 1) and validated on 2 external data sets comprising Heidelberg Spectralis SD-OCT scans (data sets 2 and 3). Main Outcomes and Measures: Prediction of progression to GA within 13 months was evaluated with area under the receiver-operator characteristic curves (AUROC) as well as area under the precision-recall curve (AUPRC), sensitivity, specificity, positive predictive value, negative predictive value, and accuracy. Results: The study included a total of 417 patients: 316 in data set 1 (mean [SD] age, 74 [8]; 185 [59%] female), 53 in data set 2, (mean [SD] age, 83 [8]; 32 [60%] female), and 48 in data set 3 (mean [SD] age, 81 [8]; 32 [67%] female). The AUROC for prediction of progression from iAMD to GA within 1 year was 0.94 (95% CI, 0.92-0.95; AUPRC, 0.90 [95% CI, 0.85-0.95]; sensitivity, 0.88 [95% CI, 0.84-0.92]; specificity, 0.90 [95% CI, 0.87-0.92]) for data set 1. The addition of expert-annotated SD-OCT features to the model resulted in no improvement compared to the fully autonomous model (AUROC, 0.95; 95% CI, 0.92-0.95; P = .19). On an independent validation data set (data set 2), the model predicted progression to GA with an AUROC of 0.94 (95% CI, 0.91-0.96; AUPRC, 0.92 [0.89-0.94]; sensitivity, 0.91 [95% CI, 0.74-0.98]; specificity, 0.80 [95% CI, 0.63-0.91]). At a high-specificity operating point, simulated clinical trial recruitment was enriched for patients progressing to GA within 1 year by 8.3- to 20.7-fold (data sets 2 and 3). Conclusions and Relevance: The fully automated, position-aware deep-learning algorithm assessed in this study successfully predicted progression from iAMD to GA over a clinically meaningful time frame. The ability to predict imminent GA progression could facilitate clinical trials aimed at preventing the condition and could guide clinical decision-making regarding screening frequency or treatment initiation.

Facilitating Harmonization of Variables in Framingham, MESA, ARIC, and REGARDS Studies Through a Metadata Repository.

BACKGROUND: High-quality research in cardiovascular prevention, as in other fields, requires inclusion of a broad range of data sets from different sources. Integrating and harmonizing different data sources are essential to increase generalizability, sample size, and representation of understudied populations-strengthening the evidence for the scientific questions being addressed. METHODS: Here, we describe an effort to build an open-access repository and interactive online portal for researchers to access the metadata and code harmonizing data from 4 well-known cohort studies-the REGARDS (Reasons for Geographic and Racial Differences in Stroke) study, FHS (Framingham Heart Study), MESA (Multi-Ethnic Study of Atherosclerosis), and ARIC (Atherosclerosis Risk in Communities) study. We introduce a methodology and a framework used for preprocessing and harmonizing variables from multiple studies. RESULTS: We provide a real-case study and step-by-step guidance to demonstrate the practical utility of our repository and interactive web page. In addition to our successful development of such an open-access repository and interactive web page, this exercise in harmonizing data from multiple cohort studies has revealed several key themes. These themes include the importance of careful preprocessing and harmonization of variables, the value of creating an open-access repository to facilitate collaboration and reproducibility, and the potential for using harmonized data to address important scientific questions and disparities in cardiovascular disease research. CONCLUSIONS: By integrating and harmonizing these large-scale cohort studies, such a repository may improve the statistical power and representation of understudied cohorts, enabling development and validation of risk prediction models, identification and investigation of risk factors, and creating a platform for racial disparities research. REGISTRATION: URL: https://precision.heart.org/duke-ninds.

Deep-Learning-Based Screening and Ancillary Testing for Thyroid Cytopathology.

Thyroid cancer is the most common malignant endocrine tumor. The key test to assess preoperative risk of malignancy is cytologic evaluation of fine-needle aspiration biopsies (FNABs). The evaluation findings can often be indeterminate, leading to unnecessary surgery for benign post-surgical diagnoses. We have developed a deep-learning algorithm to analyze thyroid FNAB whole-slide images (WSIs). We show, on the largest reported data set of thyroid FNAB WSIs, clinical-grade performance in the screening of determinate cases and indications for its use as an ancillary test to disambiguate indeterminate cases. The algorithm screened and definitively classified 45.1% (130/288) of the WSIs as either benign or malignant with risk of malignancy rates of 2.7% and 94.7%, respectively. It reduced the number of indeterminate cases (N = 108) by reclassifying 21.3% (N = 23) as benign with a resultant risk of malignancy rate of 1.8%. Similar results were reproduced using a data set of consecutive FNABs collected during an entire calendar year, achieving clinically acceptable margins of error for thyroid FNAB classification.

Differential host responses within the upper respiratory tract and peripheral blood of children and adults with SARS-CoV-2 infection.

Age is among the strongest risk factors for severe outcomes from SARS-CoV-2 infection. We sought to evaluate associations between age and both mucosal and systemic host responses to SARS-CoV-2 infection. We profiled the upper respiratory tract (URT) and peripheral blood transcriptomes of 201 participants (age range of 1 week to 83 years), including 137 non-hospitalized individuals with mild SARS-CoV-2 infection and 64 uninfected individuals. Among uninfected children and adolescents, young age was associated with upregulation of innate and adaptive immune pathways within the URT, suggesting that young children are primed to mount robust mucosal immune responses to exogeneous respiratory pathogens. SARS-CoV-2 infection was associated with broad induction of innate and adaptive immune responses within the URT of children and adolescents. Peripheral blood responses among SARS-CoV-2-infected children and adolescents were dominated by interferon pathways, while upregulation of myeloid activation, inflammatory, and coagulation pathways was observed only in adults. Systemic symptoms among SARS-CoV-2-infected subjects were associated with blunted innate and adaptive immune responses in the URT and upregulation of many of these same pathways within peripheral blood. Finally, within individuals, robust URT immune responses were correlated with decreased peripheral immune activation, suggesting that effective immune responses in the URT may promote local viral control and limit systemic immune activation and symptoms. These findings demonstrate that there are differences in immune responses to SARS-CoV-2 across the lifespan, including between young children and adolescents, and suggest that these varied host responses contribute to observed differences in the clinical presentation of SARS-CoV-2 infection by age. One Sentence Summary: Age is associated with distinct upper respiratory and peripheral blood transcriptional responses among children and adults with SARS-CoV-2 infection.

Serum Metabolites Are Associated With HFpEF in Biopsy-Proven Nonalcoholic Fatty Liver Disease.

Background Nonalcoholic fatty liver disease (NAFLD) and heart failure with preserved ejection fraction (HFpEF) share common risk factors, including obesity and diabetes. They are also thought to be mechanistically linked. The aim of this study was to define serum metabolites associated with HFpEF in a cohort of patients with biopsy-proven NAFLD to identify common mechanisms. Methods and Results We performed a retrospective, single-center study of 89 adult patients with biopsy-proven NAFLD who had transthoracic echocardiography performed for any indication. Metabolomic analysis was performed on serum using ultrahigh performance liquid and gas chromatography/tandem mass spectrometry. HFpEF was defined as ejection fraction >50% plus at least 1 echocardiographic feature of HFpEF (diastolic dysfunction, abnormal left atrial size) and at least 1 heart failure sign or symptom. We performed generalized linear models to evaluate associations between individual metabolites, NAFLD, and HFpEF. Thirty-seven out of 89 (41.6%) patients met criteria for HFpEF. A total of 1151 metabolites were detected; 656 were analyzed after exclusion of unnamed metabolites and those with >30% missing values. Fifty-three metabolites were associated with the presence of HFpEF with unadjusted P value <0.05; none met statistical significance after adjustment for multiple comparisons. The majority (39/53, 73.6%) were lipid metabolites, and levels were generally increased. Two cysteine metabolites (cysteine s-sulfate and s-methylcysteine) were present at significantly lower levels in patients with HFpEF. Conclusions We identified serum metabolites associated with HFpEF in patients with biopsy-proven NAFLD, with increased levels of multiple lipid metabolites. Lipid metabolism could be an important pathway linking HFpEF to NAFLD.

Development, Deployment, and Implementation of a Machine Learning Surgical Case Length Prediction Model and Prospective Evaluation.

OBJECTIVE: To implement a machine learning model using only the restricted data available at case creation time to predict surgical case length for multiple services at different locations. BACKGROUND: The operating room is one of the most expensive resources in a health system, estimated to cost $22 to $133 per minute and generate about 40% of hospital revenue. Accurate prediction of surgical case length is necessary for efficient scheduling and cost-effective utilization of the operating room and other resources. METHODS: We introduced a similarity cascade to capture the complexity of cases and surgeon influence on the case length and incorporated that into a gradient-boosting machine learning model. The model loss function was customized to improve the balance between over- and under-prediction of the case length. A production pipeline was created to seamlessly deploy and implement the model across our institution. RESULTS: The prospective analysis showed that the model output was gradually adopted by the schedulers and outperformed the scheduler-predicted case length from August to December 2022. In 33,815 surgical cases across outpatient and inpatient platforms, the operational implementation predicted 11.2% fewer underpredicted cases and 5.9% more cases within 20% of the actual case length compared with the schedulers and only overpredicted 5.3% more. The model assisted schedulers to predict 3.4% more cases within 20% of the actual case length and 4.3% fewer underpredicted cases. CONCLUSIONS: We created a unique framework that is being leveraged every day to predict surgical case length more accurately at case posting time and could be potentially utilized to deploy future machine learning models.

Convolutional Neural Network-Based Automated Quality Assessment of OCT and OCT Angiography Image Maps in Individuals With Neurodegenerative Disease.

Purpose: To train and test convolutional neural networks (CNNs) to automate quality assessment of optical coherence tomography (OCT) and OCT angiography (OCTA) images in patients with neurodegenerative disease. Methods: Patients with neurodegenerative disease were enrolled in the Duke Eye Multimodal Imaging in Neurodegenerative Disease Study. Image inputs were ganglion cell-inner plexiform layer (GC-IPL) thickness maps and fovea-centered 6-mm × 6-mm OCTA scans of the superficial capillary plexus (SCP). Two trained graders manually labeled all images for quality (good versus poor). Interrater reliability (IRR) of manual quality assessment was calculated for a subset of each image type. Images were split into train, validation, and test sets in a 70%/15%/15% split. An AlexNet-based CNN was trained using these labels and evaluated with area under the receiver operating characteristic (AUC) and summaries of the confusion matrix. Results: A total of 1465 GC-IPL thickness maps (1217 good and 248 poor quality) and 2689 OCTA scans of the SCP (1797 good and 892 poor quality) served as model inputs. The IRR of quality assessment agreement by two graders was 97% and 90% for the GC-IPL maps and OCTA scans, respectively. The AlexNet-based CNNs trained to assess quality of the GC-IPL images and OCTA scans achieved AUCs of 0.990 and 0.832, respectively. Conclusions: CNNs can be trained to accurately differentiate good- from poor-quality GC-IPL thickness maps and OCTA scans of the macular SCP. Translational Relevance: Since good-quality retinal images are critical for the accurate assessment of microvasculature and structure, incorporating an automated image quality sorter may obviate the need for manual image review.

Machine learning functional impairment classification with electronic health record data.

BACKGROUND: Poor functional status is a key marker of morbidity, yet is not routinely captured in clinical encounters. We developed and evaluated the accuracy of a machine learning algorithm that leveraged electronic health record (EHR) data to provide a scalable process for identification of functional impairment. METHODS: We identified a cohort of patients with an electronically captured screening measure of functional status (Older Americans Resources and Services ADL/IADL) between 2018 and 2020 (N = 6484). Patients were classified using unsupervised learning K means and t-distributed Stochastic Neighbor Embedding into normal function (NF), mild to moderate functional impairment (MFI), and severe functional impairment (SFI) states. Using 11 EHR clinical variable domains (832 variable input features), we trained an Extreme Gradient Boosting supervised machine learning algorithm to distinguish functional status states, and measured prediction accuracies. Data were randomly split into training (80%) and test (20%) sets. The SHapley Additive Explanations (SHAP) feature importance analysis was used to list the EHR features in rank order of their contribution to the outcome. RESULTS: Median age was 75.3 years, 62% female, 60% White. Patients were classified as 53% NF (n = 3453), 30% MFI (n = 1947), and 17% SFI (n = 1084). Summary of model performance for identifying functional status state (NF, MFI, SFI) was AUROC (area under the receiving operating characteristic curve) 0.92, 0.89, and 0.87, respectively. Age, falls, hospitalization, home health use, labs (e.g., albumin), comorbidities (e.g., dementia, heart failure, chronic kidney disease, chronic pain), and social determinants of health (e.g., alcohol use) were highly ranked features in predicting functional status states. CONCLUSION: A machine learning algorithm run on EHR clinical data has potential utility for differentiating functional status in the clinical setting. Through further validation and refinement, such algorithms can complement traditional screening methods and result in a population-based strategy for identifying patients with poor functional status who need additional health resources.

Enhancing early autism prediction based on electronic records using clinical narratives.

Recent work has shown that predictive models can be applied to structured electronic health record (EHR) data to stratify autism likelihood from an early age (<1 year). Integrating clinical narratives (or notes) with structured data has been shown to improve prediction performance in other clinical applications, but the added predictive value of this information in early autism prediction has not yet been explored. In this study, we aimed to enhance the performance of early autism prediction by using both structured EHR data and clinical narratives. We built models based on structured data and clinical narratives separately, and then an ensemble model that integrated both sources of data. We assessed the predictive value of these models from Duke University Health System over a 14-year span to evaluate ensemble models predicting later autism diagnosis (by age 4 years) from data collected from ages 30 to 360 days. Our sample included 11,750 children above by age 3 years (385 meeting autism diagnostic criteria). The ensemble model for autism prediction showed superior performance and at age 30 days achieved 46.8% sensitivity (95% confidence interval, CI: 22.0%, 52.9%), 28.0% positive predictive value (PPV) at high (90%) specificity (CI: 2.0%, 33.1%), and AUC4 (with at least 4-year follow-up for controls) reaching 0.769 (CI: 0.715, 0.811). Prediction by 360 days achieved 44.5% sensitivity (CI: 23.6%, 62.9%), and 13.7% PPV at high (90%) specificity (CI: 9.6%, 18.9%), and AUC4 reaching 0.797 (CI: 0.746, 0.840). Results show that incorporating clinical narratives in early autism prediction achieved promising accuracy by age 30 days, outperforming models based on structured data only. Furthermore, findings suggest that additional features learned from clinician narratives might be hypothesis generating for understanding early development in autism.